The aim of this research project is to realise an evolving Spiking Neural Networks (SNN) topology on an FPGA with on board microprocessor dedicated to running an Evolutionary Algorithm (EA), with applications in mobile robotics
Spiking neuron approaches offer a more biologically plausible neuron model. Artificial neurons have been the subject of considerable research with models ranging from the first McCulloch-Pitt binary neurons to the conventional sigmoidal neurons and then onto the contemporary spiking neurons. It is believed that these contemporary models hold a new level of computational power that has yet to be fully exploited. Spike models have been developed and a supervised off-chip learning algorithm used to extract the final weights and delays for the hardware network to solve a non-linearly separable problem. These results have led to an IEEE publication. Three future areas for research are being pursued. The first is the development of an on-chip hardware realisable supervised learning algorithm employing an evolutionary strategy. The second is the investigation into the pragmatism of the previously developed spike models to suitable benchmark problems via the developed on-chip learning algorithm. Finally it proposed to devise evolutionary techniques which will allow network connections to evolve leading to evolvable spiking neural networks. This will facilitate both adaptation and leaning in an SNN model. The results of the research are being applied to obstacle avoidance in mobile robotics.